In the wake of criticisms on the disappointing performance of its AI products, especially in areas like notification summariesApple Monday detailed How he tries to improve his AI models by analyzing the user data in private using synthetic data.
Use of an approach called “differential confidentiality“, The company said that it would first generate synthetic data, then user devices (provided they have opted to share the analysis of devices with Apple) with extracts from the synthetic data generated to compare the accuracy of its models and improve them later.
“Synthetic data is created to imitate the format and important properties of user data, but do not contain any content generated by the user,” wrote the company in a blog article. “To organize a representative set of synthetic emails, we start by creating a large set of synthetic messages on a variety of subjects (…), we then derive a representation, called integration, of each synthetic message which captures some of the key dimensions of the message, material, subject and length.”
The company said that these interests are then sent to a small number of user devices which have opted for the analysis of the devices, and the devices then compare them with a sample of emails to tell Apple which integrations are the most precise.
The company said that it used this approach to improve its Genmoji models and, in the future, would use synthetic data for image game image, image wand, creation of memories and writing tools as well as visual intelligence. Apple said it would also question users who adapt to share the analysis of devices with synthetic data to improve email summaries.